As advanced Android malware becomes increasingly sophisticated, the application of deep learning models is becoming crucial. However, powerful detection models often lead to rapid battery drain, creating a trade-off between security and energy consumption. This study tests various Multi-Layer Perceptron (MLP) model configurations to balance malware detection performance and energy efficiency. We compared standard FP32 models with optimized INT8 quantized neural networks using the TUANDROMD and DREBIN datasets to assess classification performance and energy consumption. The results show that INT8 quantization reduces model size by about 3.5 times, with energy consumption dropping to 0.0189 mJ per inference, while maintaining over 99.2% detection accuracy. We found that shallow quantized architectures, such as 3-layer and 4-layer QNNs, significantly reduce energy costs by improving throughput and shortening the time the CPU operates in a high-power state. This work demonstrates that efficient malware protection can be achieved on resource-constrained smartphones and lays the groundwork for Green AI in mobile security.
Blogger's Review: This paper effectively balances the trade-off between security and energy consumption in Android malware detection through quantization techniques, marking an important exploration of Green AI in mobile security. By optimizing model configurations, the research provides practical solutions for devices with limited resources, showcasing broad application potential.